Expression Recognition Method of Image Sequence in Audio-video

Speaking has a significant effect on facial expression recognition in a long audio-video. So it extremely affects the correct recognition rate of expression recognition with the existing methods. To overcome this disadvantage, a new method based expression recognition of audio-video is presented in this paper. In this method, audio-visual file is separated into video file and audio file first. Then the image sequences which not include speaking in video file are obtained by speech endpoint detection in audio file. After image pre-processing and feature extraction, the expression of each image is recognized by fuzzy buried Markov model classifier. At last, the recognition results of each image are fused by the hierarchical decision fusion algorithm in decision level. The experimental result shows that this method applied to the expression recognition of long audio-video can effectively reduce the negative impact of expression recognition when person is speaking, and raise the accuracy of expression recognition.